LLM 0.32 release candidates: new default model, OpenAI-compatible endpoints, and a local chat-completions server

rc1 introduces content-addressable hash IDs for stored messages — enables database dedup and branching conversation trees, via new tables only (existing logs.db data untouched, though a backup is still recommended before upgrading).
rc2 (same day) changes the default model from GPT-4o mini to GPT-5.6 Luna ($0.20/$1.20 per M tokens, vs. 4o mini’s $0.15/$0.60) and adds an llm openai endpoint command to hit arbitrary OpenAI-compatible endpoints — demoed against a local LM Studio model via uvx, no prior llm install required.
Companion plugin llm-chat-completions-server (0.1a0) uses rc1’s new hash-based logging to serve a local OpenAI-compatible Chat Completions API (default port 9001) — notably, Willison says “GPT-5.6 Sol wrote the whole thing,” calling out the model’s fluency with the OpenAI API shape.
Takeaway: a fast-moving, coordinated release across three packages in one day — schema rework, pricing-driven default-model change, and AI-authored tooling to expose it all as a drop-in local API.